Chenlu LIN, Lynn

The Chinese University of Hong Kong

• Model Distillation: Distilled GroundingSAM into YOLO for ROI extraction; integrated HRNet for high-precision heatmap and keypoint mapping.
• Inference Optimization: Implemented Kalman Filter tracking (1/5 inference frequency), reducing latency and boosting FPS on edge devices.
• Active Learning: Implemented a "Train-Infer-Correct-Retrain" loop where the HRNet model iteratively labels data, and humans only verify low-confidence samples.
• 3D Reconstruction: Built a multi-camera 3D skeleton dataset using multi-view fusion; achieved high-fidelity 3D pose estimation.

The Chinese University of Hong Kong

• Workflow Logic: Analyzed medical operational workflows to design APIs that ensure seamless information flow between 11 core modules.
• Data Management: Maintained MySQL databases to ensure the integrity and security of sensitive medical and administrative information.
• Quality Assurance: Developed unit tests to minimize logic errors and ensure data reliability.
• Performance Optimization: Resolved performance bottlenecks identified through load testing by implementing Redis caching and database indexing.

Tech: Java, Vue, SpringBoot, MyBatis, MySQL, Redis, JUnit, JMeter, Postman, JWT.

The Chinese University of Hong Kong

• Project Coordination: Managed the full lifecycle of the application, coordinating front-end and back-end developers to keep the project on schedule.
• Team Collaboration: Authored detailed technical manuals and operational workflows to streamline team operations.
• System Stability: Replaced HTTP polling with WebSocket and implemented a 15s heartbeat with exponential backoff to ensure real-time communication and prevent server congestion.
• Privacy Protection: Designed milestone-based check-ins to avoid constant GPS monitoring, balancing operational verification with user privacy.

Tech: Kotlin, Python, Azure, FastAPI, MongoDB, JWT, WebSocket, Postman, JMeter, Figma.

Fuzhou University

• Flight Control Adjustment: Implemented flight correction and stabilization using GPS applications combined with gravity and optical flow sensors.
• Pose Estimation Algorithm: Trained a model based on the HRNet network and deployed it on a small computing device carried by the UAV to achieve human pose estimation.